Grant: Discovery Grants Program - Individual

The brain regulates emotion, so signals from the brain (EEG) contain emotional information. These EEG signals can help to train machine learning (ML) models to recognize emotions. However, because EEG signals vary among individuals, the accuracy of these models can drop for new users, needing recalibration. This makes the models less usable. Most prior research has addressed this through machine learning techniques, like feature selection and transfer learning. Instead of focusing on that direction, we plan to use neuroscience theories. There might be emotion processing patterns shared across individuals, so we will explore this and apply it to build ML models that work well for a broad population. The long-term goal of my research program is to develop methods to extract emotional content from physiological signals to build generalizable emotion recognition approaches. To achieve this, in the next five years, we will use a new model of emotional perception. This model suggests that, initially, the right hemisphere is more active in emotion perception, but after exposure to a stimulus, the left hemisphere also becomes activated. The activation response in the left hemisphere depends on the stimulus type, and we will use these distinct activation patterns to boost emotion recognition. The short-term objectives are: 1. Perform an analytical comparison of hemisphere activation differences evoked by emotional stimuli. This will identify emotion-associated activation patterns. 2. Develop an ensemble approach that trains predictive models over the activation differences between the hemispheres. This will extract hemispheric activation variation to improve emotion prediction. 3. Develop a module that identifies the relevant features for emotion processing. This will recognize those features that can discriminate emotions across different individuals. 4. Develop deep learning approaches that use information that captures simultaneous hemispheric activation across individuals. This will extract activation difference patterns that facilitate emotion recognition. This research program advances EEG emotion recognition by introducing a novel neuroscience model. So, it bridges theories from neuroscience with data-driven techniques. This connection helps identify neuronal patterns related to emotion processing. As these patterns are shared across individuals, they enable the creation of generalizable emotion recognition systems, reducing recalibration needs for new users. Also, these systems promote inclusivity and reduce ethical biases. In our AI-driven society, our research contributes to Canadians by providing methods to build emotion prediction models that work well for a broad population. These models are relevant in diverse domains such as human-computer interaction, healthcare, education, and marketing. For instance, they can help manage conditions like anxiety, depression, and bipolar disorders, mainly affecting young people and females in Canada.